Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76632.52307124846,
-      "required_pages_per_second_upper_estimate": 0.0053016157580202
+      "required_token_positions_per_second": 77375.31608524587,
+      "required_pages_per_second_upper_estimate": 0.005353003902245787
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9579.065383906058,
-      "required_pages_per_second_upper_estimate": 0.004010554642609668
+      "required_token_positions_per_second": 9671.914510655733,
+      "required_pages_per_second_upper_estimate": 0.00404942863306951
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19158.130767812116,
-      "required_pages_per_second_upper_estimate": 0.003990133405051409
+      "required_token_positions_per_second": 19343.829021311467,
+      "required_pages_per_second_upper_estimate": 0.0040288094540630575
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28737.196151718177,
-      "required_pages_per_second_upper_estimate": 0.0017227510301376667
+      "required_token_positions_per_second": 29015.743531967197,
+      "required_pages_per_second_upper_estimate": 0.0017394495202663724
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28737.196151718177,
-      "required_pages_per_second_upper_estimate": 0.002234190599228746
+      "required_token_positions_per_second": 29015.743531967197,
+      "required_pages_per_second_upper_estimate": 0.002255846432842665
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28737.196151718177,
-      "required_pages_per_second_upper_estimate": 0.3686098966370131
+      "required_token_positions_per_second": 29015.743531967197,
+      "required_pages_per_second_upper_estimate": 0.3721828033499724
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1899474762265072,
-    "producer_p95_seconds": 0.33408611800405197,
-    "gpu_update_seconds": 0.9620980367754822,
-    "producer_batches_per_second": 5.264613249230683,
-    "gpu_batches_per_second": 1.0393951154411956,
-    "producer_headroom": 5.065074071467032,
+    "producer_batch_seconds": 0.21827482950027388,
+    "producer_p95_seconds": 0.34984496499964735,
+    "gpu_update_seconds": 0.9528620202180816,
+    "producer_batches_per_second": 4.58138028232314,
+    "gpu_batches_per_second": 1.0494698904791377,
+    "producer_headroom": 4.3654232712017125,
     "sustainable": true,
-    "buffer_seconds": 60.61217631685538,
+    "buffer_seconds": 60.03030727373914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 74,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381656.116714488,
+    "generation_bytes_per_second": 399858.9482914248,
     "upload_bytes_per_second": 3787701.316098853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.8945181472395,
+    "estimated_full_shards_per_hour": 85.80042206341798,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 524600, "loss": 0.5592507123947144, "lr": 0.0003, "grad_norm": 0.15388680595723428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76783892664, "tokens_per_second": 154076.70086355964, "data_wait_s": 1.5998994058463722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524700, "loss": 0.5254177451133728, "lr": 0.0003, "grad_norm": 0.17957341772498586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76798479936, "tokens_per_second": 151887.0193837863, "data_wait_s": 1.7935999494511634e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524800, "loss": 0.5686387419700623, "lr": 0.0003, "grad_norm": 0.15876198911507197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76812983798, "tokens_per_second": 151723.5168645376, "data_wait_s": 1.5948993677739054e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 524900, "loss": 0.5441434383392334, "lr": 0.0003, "grad_norm": 0.198230476678356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76827643313, "tokens_per_second": 152883.9950594783, "data_wait_s": 1.6022997442632914e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d75575e09550f81fe5dfe049d7bc402d3b26d8811ed974a89c4f8616c6697c82
-size 848610
+oid sha256:dc0348ec70f6cf9715a4811a0c91f5c509603ab6eb22312a4792fdf4ee970537
+size 850728